Perl - 243
(Not including linebreaks, all of which are optional)
Mostly de-golfed code can be found here.
$/="";@w=split//,pop;($b=<>)=~y/ //d;$W=1+index$b,"\n";
for(1,2){for(0,$W-2..$W){$p=join"."x$_,@w;
push@m,[$-[0],$+[0]-1]while$b=~/$p/sg}
@w=reverse@w;@m=map[reverse@$_],@m}
$t="row %d, column %d";
printf"$t --> $t\n",map{1+$_/$W,1+$_%$W}@$_ for@m;
Usage: run as a script. Either wordsearch.pl boardfile searchword to read from a file, or wordsearch.pl searchword to read the board from STDIN (hey, pop is two characters shorter than shift!)
The idea is that we read the board into a string (throwing away the extra spaces between letters), and also make note of how wide it is (by counting characters until the first newline). Then we do a series of regex matches to find the word, using the s regex modifier to allow a match to span lines. Given a board that's 4 letters wide (so each line is 5 chars including newline), and the search word "CAR", we can make the patterns /CAR/ to match forwards, /C...A...R/ to match going southwest, /C....A....R/ to match going downwards, and /C.....A.....R/ to match going southeast. For each match, we record the starting and ending positions of the match.
Then we reverse the search word ("CAR" -> "RAC") and do it again, which lets us match the word going left, up, northwest, and northeast. Of course, we want to make sure that the start/end positions are also swapped. To save code, I swap the match positions twice; the matches for the forward word are swapped both times (so they come out un-swapped), while the matches for the backward word are only swapped once, so they will come out swapped.
Finally, it's just a matter of taking the match offsets into the board string and turning them into row/column pairs, and formatting the output how the problem wants it.
